具有两个条件重置的时间序列累积和[重复]答案

【问题标题】：Time series cumulative sum with two conditional resets [duplicate]具有两个条件重置的时间序列累积和[重复]
【发布时间】：2021-10-02 21:28:10
【问题描述】：

我正在寻找一种（最好是 tidyverse）解决方案来创建基于列 direction 的运行计数器（累积和），每次方向更改或停止时都会重置 counter。

library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union
library(lubridate)
#> 
#> Attaching package: 'lubridate'
#> The following objects are masked from 'package:base':
#> 
#>     date, intersect, setdiff, union

df <- tibble::tribble(
    ~direction,             ~datetime, ~counter,
        "Left",  "2020-09-27 8:00:00",       1L,
        "Left",  "2020-09-27 9:00:00",       2L,
       "Right", "2020-09-27 10:00:00",       1L,
       "Right", "2020-09-27 11:00:00",       2L,
       "Right", "2020-09-27 12:00:00",       3L,
        "Stop", "2020-09-27 13:00:00",       1L,
        "Left", "2020-09-27 14:00:00",       1L,
       "Right", "2020-09-27 15:00:00",       1L,
        "Left", "2020-09-27 16:00:00",       1L,
        "Left", "2020-09-27 17:00:00",       2L
    ) %>% 
  mutate(datetime = as_datetime(datetime))
df
#> # A tibble: 10 x 3
#>    direction datetime            counter
#>    <chr>     <dttm>                <int>
#>  1 Left      2020-09-27 08:00:00       1
#>  2 Left      2020-09-27 09:00:00       2
#>  3 Right     2020-09-27 10:00:00       1
#>  4 Right     2020-09-27 11:00:00       2
#>  5 Right     2020-09-27 12:00:00       3
#>  6 Stop      2020-09-27 13:00:00       1
#>  7 Left      2020-09-27 14:00:00       1
#>  8 Right     2020-09-27 15:00:00       1
#>  9 Left      2020-09-27 16:00:00       1
#> 10 Left      2020-09-27 17:00:00       2

^{由reprex package (v2.0.0) 于 2021-07-27 创建}

我尝试调整来自 this solution 的代码，但无法让它在两次条件重置时工作。

【问题讨论】：

标签： r dplyr

【解决方案1】：

在基础 R 中，您可以使用 rle 和 sequence -

df$counter <- sequence(rle(df$direction)$lengths)

如果你想在管道中使用它，你可以这样做 -

library(dplyr)
df %>% mutate(counter = sequence(rle(direction)$lengths))

#  direction datetime            counter
#   <chr>     <dttm>                <int>
# 1 Left      2020-09-27 08:00:00       1
# 2 Left      2020-09-27 09:00:00       2
# 3 Right     2020-09-27 10:00:00       1
# 4 Right     2020-09-27 11:00:00       2
# 5 Right     2020-09-27 12:00:00       3
# 6 Stop      2020-09-27 13:00:00       1
# 7 Left      2020-09-27 14:00:00       1
# 8 Right     2020-09-27 15:00:00       1
# 9 Left      2020-09-27 16:00:00       1
#10 Left      2020-09-27 17:00:00       2

【讨论】：

【解决方案2】：

我们可以使用rowid 和rleid

library(dplyr)
library(data.table)
df %>% 
    mutate(counter2 = rowid(rleid(direction)))

-输出

# A tibble: 10 x 4
   direction datetime            counter counter2
   <chr>     <dttm>                <int>    <int>
 1 Left      2020-09-27 08:00:00       1        1
 2 Left      2020-09-27 09:00:00       2        2
 3 Right     2020-09-27 10:00:00       1        1
 4 Right     2020-09-27 11:00:00       2        2
 5 Right     2020-09-27 12:00:00       3        3
 6 Stop      2020-09-27 13:00:00       1        1
 7 Left      2020-09-27 14:00:00       1        1
 8 Right     2020-09-27 15:00:00       1        1
 9 Left      2020-09-27 16:00:00       1        1
10 Left      2020-09-27 17:00:00       2        2

或者只使用dplyr

df %>% 
    group_by(grp = cumsum(direction != 
        lag(direction, default = first(direction)))) %>% 
    mutate(counter2 = row_number()) %>%
    ungroup %>% 
    select(-grp)
# A tibble: 10 x 4
   direction datetime            counter counter2
   <chr>     <dttm>                <int>    <int>
 1 Left      2020-09-27 08:00:00       1        1
 2 Left      2020-09-27 09:00:00       2        2
 3 Right     2020-09-27 10:00:00       1        1
 4 Right     2020-09-27 11:00:00       2        2
 5 Right     2020-09-27 12:00:00       3        3
 6 Stop      2020-09-27 13:00:00       1        1
 7 Left      2020-09-27 14:00:00       1        1
 8 Right     2020-09-27 15:00:00       1        1
 9 Left      2020-09-27 16:00:00       1        1
10 Left      2020-09-27 17:00:00       2        2

【讨论】：